Log in
Register|
john john ☆ Spain, 2014-09-26 11:16 (4285 d 14:18 ago) Posting: # 13589 Views: 7,395 |
|
|
John John We had made a bioequivalence with more than 70 subjects divided into 7 groups. The statistical analysis has been like a normal bioequivalence. Regulatory agency asks us to include the covariable "group". Is this normal? From which number of groups/subjects would there be necessary to include it? It woul be possible to justify doesn't to do it? Thanks |
|
ElMaestro ★★★ Denmark, 2014-09-26 12:05 (4285 d 13:30 ago) @ john john Posting: # 13590 Views: 6,302 |
|
|
Hello John John, ❝ We had made a bioequivalence with more than 70 subjects divided into 7 groups. The statistical analysis has been like a normal bioequivalence. Regulatory agency asks us to include the covariable "group". ❝ Is this normal? That's a fairly decent request. I guess if anything is not normal then it's probably your decision to divide into 7 groups. Anyways, including group as a fixed factor (not a covariate or random factor) takes your partitioning into consideration. ❝ From which number of groups/subjects would there be necessary to include it? ❝ It woul be possible to justify doesn't to do it? When you specify group as a fixed factor the software should take care of the rest. Your model is then ln(PK)=Subject+Sequence+Group+Period+Treatment+error (at least if it is a crossover design; someone will probably suggest to make the model stuff overly complicated by using heavy nesting like subject in sequence in group etc - that is also correct but not really necessary for most practical purposes) You will see up to 7 df's for Group depending on the type of SS you wish and your specific software but df=7 is only if you do type I SS without intercept and have Group specified lexically as the first factor. Group is a between-factor. It will not affect the GMR and it will not increase the width of the CI. So the regulatory request is not a danger to the conclusion of BE, I believe. I would not know how to justify not to do it when the agency specifically asks for it, and I do not see a reason to argue against it. In EU a relevant ref. is the requirement "The statistical analysis should take into account sources of variation that can be reasonably assumed to have an effect on the response variable" - that's simply the thinking behind the request from your regulator in this casee too, whether they are European or not. Best regards, ElMaestro ElMaestro  — Pass or fail! ElMaestro |
|
john john ☆ Spain, 2014-09-26 15:01 (4285 d 10:33 ago) @ ElMaestro Posting: # 13592 Views: 6,227 |
|
|
ElMaestro ★★★ Denmark, 2014-09-26 15:13 (4285 d 10:21 ago) @ john john Posting: # 13593 Views: 6,210 |
|
|
Hello John John, ❝ Some authors think that would apply only when the intra-subject variance is high and sample size is small. Which authors? About what exactly are we really talking about now? Assuming this is related to the 7 groups discussed in another thread earlier today I believe high intra-CV, small sample size and the requirement for group as a term in the model fit are unrelated. — Pass or fail! ElMaestro |
|
jag009 ★★★ NJ, 2014-10-01 23:04 (4280 d 02:31 ago) @ john john Posting: # 13641 Views: 6,111 |
|
|
Hi, ❝ Some authors think that would apply only when the intra-subject variance is high and sample size is small. That has nothing to do with your situation. Since you ran a BE study with multiple groups, you need to demonstrate the absence of group effect. If there is a group effect then you would have to analyze the groups separately ⇒ This will elevate your CV (probably?) and reduce your power due to smaller sample size per group. I doubt that you will see a group effect if you ran the study groups in the same clinic and in different but within reasonable time frame. See previous posts about group effect etc. There were some postings(I made some too) before... John |
|
Ben ★ Germany, 2014-10-05 12:57 (4276 d 12:38 ago) @ jag009 Posting: # 13651 Views: 5,964 |
|
|
Dear All, This may be of interest: Under VII, A. the FDA guidance "Statistical Approaches Establishing BE" (2001) states "If a crossover study is carried out in two or more groups of subjects (e.g., if for logistical reasons only a limited number of subjects can be studied at one time), the statistical model should be modified to reflect the multigroup nature of the study. ..." I believe this is where they are coming from. Best, Ben |
Ing. Helmut Schütz